Nvidia has just started a new initiative with PGI to help techies speed up their scientific or industrial code by using GPU directives.
The OpenACC directive page has just been launched here. Nvidia tells us that scientists can register on the page and learn how to speed up their code with GPU number crunching.
GPU directives accelerate code by inserting compiler hints into the code and the compiler will automatically map compute-intensive portions of it to the GPU.
Obviously, Nvidia has an interest in getting people to use its GPUs to process data in parallel to standard CPUs.
But a lot of software developers find writing code to do it fairly tricky. GPUs perform best when the hundreds of cores are kept busy with thousands or even millions of threads of parallel execution.
Nvidia has given several examples of how tweaking code helped researchers at the University of Groningen, who were trying to understand the origins of the universe, speed up their program by 5.6 times.
Another case involved Global Oil Company which sped up a simulation of petrol reservoirs by three times.